Tom Lee has forecasted that more than 90% of AI companies will fail in the coming decade, drawing parallels to the early 2000s Internet bubble. He suggests that, similar to how only a few companies like Amazon and Google emerged as successful investments post-bubble, the AI sector will see a small percentage of firms delivering substantial returns. Lee advises investors to adopt an index approach, covering the entire AI ecosystem, to capitalize on the potential explosive returns from the top-performing companies while mitigating risks from widespread failures.